Skip to main content

Agent Skills

A tool gives an agent new abilities; a skill gives it new technique. Skills are instruction documents in the industry-standard Agent Skills format (a folder with a SKILL.md) that teach an agent how to do something well: a review checklist, a video-editing workflow, a theming system. Install one and every agent you enable it on works that way.

Where skills come from​

  • The community skills catalog — installable skill packages, browsed right from the dashboard (Browse Community Skills, from the admin Skills page or any agent's Skills tab). Installing a package is an admin action — a manager can request one, and an admin's approval installs it. Updates arrive with the same weekly auto-update that keeps community tools current.
  • Bundled with a tool — many MCPs ship their own how-to skills; those ride the tool's enablement automatically and don't need separate management.
  • Your own ZIP — an admin can install a skill package from a ZIP on the admin Skills page. Both layouts are accepted: a full community-format package (with a manifest.json), or a bare, spec-standard folder containing just a SKILL.md and its reference files. OtoDock fills in the packaging for you.

Skills can bundle reference files (templates, examples, data) and executable scripts the agent may run as part of the technique. Packages that do are labeled with a "bundles scripts" badge in the catalog and admin views, so you can see at a glance which packages ship runnable code. Scripts run inside the agent's ordinary sandbox like anything else the agent executes; the badge is about provenance, not extra power. On a remote machine, an agent's skills travel with it — they are copied to the machine and load there the same way, scripts included.

Two loading modes​

Every skill loads in one of two ways, shown as a badge wherever skills appear:

  • Always in context — the full skill text sits in the agent's system prompt, in every session. For short, behavior-shaping guidance the agent should never forget.
  • On demand — the agent sees only the skill's name and description, and reads the full body the moment a task matches. This is the default, and the right mode for anything substantial: it costs nothing until it's needed.

Enabling skills per agent​

Each agent has a Skills tab (Agent Settings → Skills, visible to the agent's managers): installed skill packages appear with per-skill toggles, and the skills bundled by the agent's tools are listed alongside for reference. Enable exactly the technique each agent needs — a writing agent gets the style guide, the video agent gets the editing workflow.

Agents can also ask for skills themselves. In chat, an agent can browse the skills catalog, enable a skill directly when its manager asks and the package is already installed, or file a request an admin approves — the same request queue used for tools, with a Skill badge on the row (Admin → MCP Requests). Community agent templates that bundle skills work the same way: when a non-admin installs one, the agent is created immediately and its skill installs are queued for admin approval rather than failing.

Writing your own​

A skill is a folder:

my-skill/
  SKILL.md          # the technique itself
  reference/…       # optional supporting files, read on demand
  scripts/…         # optional executable helpers

SKILL.md starts with two lines of YAML frontmatter: a name (lowercase, hyphens) and a description. The description doubles as the on-demand activation trigger, so write what the skill does and when to use it. Everything after the frontmatter is the instruction body. Keep it focused, and push bulk reference material into separate files the agent reads when needed.

Zip the folder and install it from the admin Skills page, or contribute it to the community skills catalog for everyone.

Managing installed packages​

The admin Skills page lists every installed package with its version, how many agents have it enabled, and per-package Update / Delete. Deleting a package removes it and its per-agent enablement everywhere.

Next steps​